Risk-based machine learning classifier

Abstract

The present disclosure relates generally to a risk-based fraud identification and risk analysis system. For example, the system may receive application data from a first borrower user, determine a segment associated with the application data, apply application data to one or more machine learning (ML) models, and receive a score based at least in part upon output of the ML model.

What is claimed is: 
     
       1. A method for computing an application score for a loan, the method comprising:
 receiving, by a computer system, application data identifying a plurality of attributes of: a first borrower, and of collateral; 
 generating, by the computer system, a plurality of input features based upon the application data, wherein at least one of the plurality of input features are generated based on at least one attribute of the collateral; 
 applying the plurality of input features to a trained machine learning (ML) model to generate an output; 
 determining an application score based on the output of the trained ML model; 
 grouping each of the plurality of input features into one of a plurality of factor groups, wherein each of the plurality of factor groups is related to a different type of fraud; 
 determining a reason for how each of the input features in a factor group affects the application score; 
 determining, by the computer system, one or more actions for the application data based at least in part on the application score, the plurality of factor groups, and the determined reason; and 
 providing, by the computer system, the application score and the one or more actions to a dealer user device or to a lender user device. 
 
     
     
       2. The method of  claim 1 , wherein the application data is received from a borrower user device. 
     
     
       3. The method of  claim 1 , wherein the application data is received from a borrower user device and from the dealer user device. 
     
     
       4. The method of  claim 3 , wherein the plurality of attributes of collateral are received in application data from the dealer user device. 
     
     
       5. The method of  claim 1 , wherein the at least one of the one or more input features generated based on the at least one attribute of the collateral is generated based on a value discrepancy between a value of collateral and a loan amount. 
     
     
       6. The method of  claim 5 , wherein the collateral comprises an automobile. 
     
     
       7. The method of  claim 6 , wherein the application data identifies a value of the automobile. 
     
     
       8. The method of  claim 7 , further comprising: identifying an expected value for the automobile based on attributes of the automobile; and comparing the expected value for the automobile with the identified value of the automobile in the application data. 
     
     
       9. The method of  claim 8 , further comprising generating a feature characterizing the comparison of the expected value for the automobile with the identified value of the automobile in the application data. 
     
     
       10. The method of  claim 9 , further comprising: determining a second action based on the comparison of the expected value for the automobile with the identified value of the automobile in the application data; and providing the second action to the dealer user device or to the lender user device. 
     
     
       11. A system comprising:
 one or more processors; and 
 a non-transitory computer-readable medium including instructions that, when executed by the one or more processors, cause the one or more processors to: 
 receive application data identifying a plurality of attributes of: a first borrower, and of collateral; 
 generate a plurality of input features based upon the application data, wherein at least one of the a plurality of input features are generated based on at least one attribute of the collateral; 
 apply the a plurality of input features to a trained machine learning (ML) model to generate an output; 
 determine an application score based on the output of the trained ML model; 
 group each of the plurality of input features into one of a plurality of factor groups, wherein each of the plurality of factor groups is related to a different type of fraud; 
 determine a reason for how each of the input features in a factor group affects the application score; 
 determine one or more actions for the application data based at least in part on the application score, the plurality of factor groups, and the determined reason; and 
 provide the application score and the one or more actions to a dealer user device or to a lender user device. 
 
     
     
       12. The system of  claim 11 , wherein the application data is received from a borrower user device. 
     
     
       13. The system of  claim 11 , wherein the application data is received from a borrower user device and from the dealer user device. 
     
     
       14. The system of  claim 13 , wherein the plurality of attributes of collateral are received in application data from the dealer user device. 
     
     
       15. The system of  claim 11 , wherein the at least one of the one or more input features generated based on the at least one attribute of the collateral is generated based on a value discrepancy between a value of collateral and a loan amount. 
     
     
       16. The system of  claim 15 , wherein the collateral comprises an automobile. 
     
     
       17. The system of  claim 16 , wherein the application data identifies a value of the automobile. 
     
     
       18. The system of  claim 17 , wherein the non-transitory computer-readable medium further includes instructions that, when executed by the one or more processors, cause the one or more processors to: identify an expected value for the automobile based on attributes of the automobile; and compare the expected value for the automobile with the identified value of the automobile in the application data. 
     
     
       19. The system of  claim 18 , wherein the non-transitory computer-readable medium further includes instructions that, when executed by the one or more processors, cause the one or more processors to generate a feature characterizing the comparison of the expected value for the automobile with the identified value of the automobile in the application data. 
     
     
       20. The system of  claim 19 , wherein the non-transitory computer-readable medium further includes instructions that, when executed by the one or more processors, cause the one or more processors to: determine a second action based on the comparison of the expected value for the automobile with the identified value of the automobile in the application data; and provide the second action to the dealer user device or to the lender user device.